| Sumario: | Social scientists have often been warned about the dangers involved in extending scale results over time and to other populations. In this article, authors discuss the problems of extending inferences beyond the content of the items tested. They conclude that, at least until more is known about sampling error, no inferences about scalability should be extended to untested items. The operations involved in defining and sampling an attitude area differ from those involved in testing a set of responses for scalability. Yet, after the attitude area is accepted as scalable and the scale types are used in the analysis of data, the definition and sampling of the area become crucial in making interpretations. The purposes of this article are, to describe the defining and item construction operations used in a role-conception study, to present the results obtained by the approach and to illustrate a position on the problem of how far the inference of scalability can be generalized from particular items.
|
|---|